https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67718
Ramana Radhakrishnan <ramana at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |WAITING
Last reconfirmed| |2015-09-30
CC| |ramana at gcc dot gnu.org
Ever confirmed|0 |1
--- Comment #3 from Ramana Radhakrishnan <ramana at gcc dot gnu.org> ---
With the latest build for aarch64-none-linux-gnu of 4.9 and trunk that I had
lying around, I see with -O2 and with and without a variety of tuning options,
which looks a whole load better than the original bug report.
copysignl:
fmov x2, d0
fmov x1, v1.d[1]
fmov x3, v0.d[1]
fmov d0, x2
lsr x1, x1, 63
bfi x3, x1, 63, 1
fmov v0.d[1], x3
ret
.size copysignl, .-copysignl
.ident "GCC: 4.9.4 20150924 (prerelease)"
I am traveling currently and can't dig any further but I remember something
being fixed in this pattern with respect to early clobbers- Aha PR61633 may be
your friend.
Can you check if you have the fix for PR61633 in your 4.9 based tree ?
In the future please report with command line options and the exact upstream
source revision used to create the issue.
regards
Ramana
